Active Vision Based Robot Navigation System
This project was funded by the National High Technology Program of China.
The system consists of a pan-tilt-translation camera platform, an omni-directional
vehicle and a Sunsparc10 workstation. The camera platform has two CCD cameras,
each one of which has five degrees of freedom. Based on only the images
from the two cameras, the vehicle moves to a predefined destination in
a cluttered laboratory environment. A real navigation system must at least
has the following units: modelling, camera calibration path planning, robot
self-location, (static and moving) obstacle detection and avoidance, local
re-planning and re-modelling. However, this project concerns only 3 of
these units which are considered closely related to the vision research.
These 3 units are:
-
Camera calibration: We thoroughly studied previous methods in the
literature and proposed an Active Vision Based Camera calibration method
[1], a Non-Parametric Calibration approach [2] and a Two Planes method[3].
-
Robot Self-Location: The window frames in this system are used as
landmarks to determine robot's position and orientation. The process was
carried out in two stages: Firstly, four straight lines are extracted by
an RHT-like approach. Secondly, based on the recorded approximative information
of the robot's odometers, a fuzzy approach was used to match the window
frame to its stored model[4].
-
Obstacle detection and avoidance: The principles of the used method
in our system are similar to those proposed by Mallot et al.in [5]. i.e.,
by re-projecting the two images from a well calibrated stereo rim onto
the ground, and comparing the two reprojected images, if the two images
are not identical, which means there are some things above the ground and
can be considered as potential obstacles of the robot.
-
[1] S. D. Ma, An Self Camera Calibration Method for Active Vision System,
IEEE-T on Robotics and Automation, Vol.2, 1996.
-
[2] M.L.Qiu and S.D.Ma, "A non-parametric approach for camera calibration",
in proc. ICCV 1995, MIT, USA, pp.224-229, 1995.
-
[3] G. Q. Wei and S. D. Ma, "Implicit and explicit camera calibration:
Theory and experiments", IEEE-T PAMI 16, No.5, 1995.
-
[4] L. Zhao, Fuzzy Theory Application in Robot Self-Location, M.S. thesis,
National Laboratory of Pattern Recognition, Institute of Automation, Chinese
Academy of Sciences, 1995.
-
[5] H.A.Mallot, H. H. Bulthoff, J.J.Little, and S. Bohrer, Inverse Perspective
Mapping Simplifies Optical Flow Computation and Obstacle Detection, Biological.
Cybernetics 64, pp.177-185, 1991.
-
The Graphical User Interface of the System
(A software written by me during the research work. For a bigger image,
click the image below.)
Return to Projects' Web Page